Synthetic heparan sulfate standards and machine learning facilitate the development of solid-state nanopore analysis

Abstract

The application of solid-state (SS) nanopore devices to single-molecule nucleic acid sequencing has been challenging. Thus, the early successes in applying SS nanopore devices to the more difficult class of biopolymer, glycosaminoglycans (GAGs), have been surprising, motivating us to examine the potential use of an SS nanopore to analyze synthetic heparan sulfate GAG chains of controlled composition and sequence prepared through a promising, recently developed chemoenzymatic route. A minimal representation of the nanopore data, using only signal magnitude and duration, revealed, by eye and image recognition algorithms, clear differences between the signals generated by four synthetic GAGs. By subsequent machine learning, it was possible to determine disaccharide and even monosaccharide composition of these four synthetic GAGs using as few as 500 events, corresponding to a zeptomole of sample. These data suggest that ultrasensitive GAG analysis may be possible using SS nanopore detection and well-characterized molecular training sets.

A Guide to Signal Processing Algorithms for Nanopore Sensors

“… 49 Other classification and clustering algorithms have also been implemented to identify various analytes via the translocation features obtained from nanopore sensors, such as the k -nearest neighbor ( k ), 89 , 90 the logistic regression, 69 , 89 , 90 and the naive Bayes. 89 …”

“…For instance, by utilizing ML, it is possible to determine two different compositions of four synthetic biopolymers using as few as 500 events. 89 Seven ML algorithms are compared: (i) AdaBoost; (ii) k -NN; (iii) naive Bayes; (iv) NN; (v) RF; (vi) logistic regression; and (vii) SVM. A minimal representation of the nanopore data, using only signal amplitude and duration, can reveal, by eye and image recognition algorithms, clear differences among the signals generated by the four glycosaminoglycans.…”

“…In addition, the same algorithm has been used to classify two viruses, influenza A and B, through the translocation signals from peptide functionalized SiN x nanopore sensors. 49 Other classification and clustering algorithms have also been implemented to identify various analytes via the translocation features obtained from nanopore sensors, such as the k-nearest neighbor (k), 89,90 the logistic regression, 69,89,90 and the naive Bayes. 89 An important facet of the ML related algorithms is that they devote significant effort to comparisons among as many methods as possible.…”

“…49 Other classification and clustering algorithms have also been implemented to identify various analytes via the translocation features obtained from nanopore sensors, such as the k-nearest neighbor (k), 89,90 the logistic regression, 69,89,90 and the naive Bayes. 89 An important facet of the ML related algorithms is that they devote significant effort to comparisons among as many methods as possible. For instance, by utilizing ML, it is possible to determine two different compositions of four synthetic biopolymers using as few as 500 events.…”

“…An important facet of the ML related algorithms is that they devote significant effort to comparisons among as many methods as possible. For instance, by utilizing ML, it is possible to determine two different compositions of four synthetic biopolymers using as few as 500 events . Seven ML algorithms are compared: (i) AdaBoost; (ii) k -NN; (iii) naive Bayes; (iv) NN; (v) RF; (vi) logistic regression; and (vii) SVM.…”